Data centers expected to use 4x more electricity by 2035

Overview

Data centers are projected to consume four times more electricity by 2035, reaching one-fifth of all U.S. electricity generation, driven primarily by AI computing demands. New data centers built through 2033 could consume as much power as India uses annually. The surge is straining electrical grids, particularly in regions like PJM and ERCOT, where capacity constraints are already pushing prices higher and forcing utility companies to reconsider their commitments.

Key Takeaways

Data center electricity consumption is expected to quadruple by 2035, accounting for one-fifth of all U.S. electricity generation.
BloombergNEF's 2035 forecast is 83% higher than its December prediction, reflecting accelerating AI adoption and development timelines.
The U.S. will host 64% of global AI chips by power demand by 2033, with most new capacity concentrated in already-strained regional grids.
The PJM Interconnection faces severe congestion, with data centers potentially consuming 34% of its electricity and causing a 76% spike in prices over the past year.
Globally, data centers could create nearly 1,935 terawatt-hours of new electricity demand by 2033, equivalent to India's current annual consumption.

Stats & Key Facts

#Data center capacity expected to reach nearly 200 gigawatts by 2035
#Data centers projected to use one-fifth of U.S. electricity by 2035, compared to one-twentieth today (a 4x increase)
#BloombergNEF's 2035 forecast is 83% higher than December prediction
#U.S. will host 64% of global AI chip capacity by power demand by 2033
#PJM electricity prices rose 76% over the past year due to grid strain
#Data centers represented 38% of charges in PJM's most recent capacity auction

Explosive Growth in Data Center Electricity Demand

The surge in artificial intelligence computing is driving unprecedented growth in data center energy consumption.

›Data centers are expected to consume four times more electricity by 2035 than they do today.
›Nearly 200 gigawatts of new data center capacity will be deployed over the next decade.
›BloombergNEF's forecast increased 83% from December, indicating forecasters have consistently underestimated demand growth.
›Multiple forecasting organizations, including EPRI and S&P, have also revised their estimates significantly upward in recent months.

The projected electricity demand from data centers reflects the accelerating pace of AI development and deployment. More than half of the new capacity will be dedicated to AI training and inference, positioning artificial intelligence as the primary driver of energy consumption growth. This concentration of AI workloads represents a fundamental shift in how the tech industry consumes power, with implications for grid infrastructure, energy policy, and climate goals.

The conservative nature of these forecasts is evident in how quickly they become outdated. When BloombergNEF released its December forecast, analysts did not anticipate the rapid scaling decisions made by major technology companies. By the time the new report was published, actual data center development plans exceeded previous projections, forcing a substantial upward revision within just a few months.

U.S. Dominance in AI Compute Infrastructure

The United States will remain the global center for AI computing, hosting a disproportionate share of data center capacity.

›By 2033, the U.S. will account for 64% of global AI chip capacity measured by power demand.
›Nearly half of new data center capacity will focus on AI training and inference workloads.
›Most of the concentrated AI compute infrastructure will remain in the United States throughout the forecast period.
›This concentration reflects existing advantages in energy infrastructure, technical talent, and venture capital investment.

The geographic concentration of AI compute in the U.S. creates both opportunities and challenges. While American data centers will benefit from first-mover advantages and easier access to capital, the concentration also places enormous stress on specific regions. The decision by major AI companies to build data centers near existing power infrastructure has created bottlenecks in interconnection-rich areas, particularly where multiple technology companies are competing for limited grid capacity.

This American dominance reflects strategic choices by major technology companies to remain close to research institutions, talent pools, and existing cloud infrastructure. However, the concentration may not persist indefinitely, as other countries develop their own AI capabilities and energy infrastructure. International competition for AI leadership will likely drive distributed data center development globally over time.

Grid Strain and Regional Capacity Challenges

Existing electrical grids are struggling to accommodate the projected surge in data center connections.

›PJM Interconnection will dedicate 34% of its electricity to data centers, the highest regional impact.
›ERCOT in Texas will allocate 22% of its generating capacity to data centers.
›PJM paused new grid connection applications for four years before reopening in April, creating severe backlogs.
›Electricity prices in PJM increased 76% over the past year as congestion worsened.
›American Electric Power has threatened to leave PJM due to interconnection challenges.

The PJM Interconnection, which serves Virginia to Illinois, represents one of the most stressed electrical grids in the country. The region already hosts a large concentration of data centers, and the continued demand for connections has created a crisis. When PJM paused applications for four years, the queue of pending requests grew dramatically, leaving utilities uncertain about future capacity and investment needs.

The reopening of PJM's queue in April did not immediately resolve the underlying problem. Even with the ability to accept new applications again, the region's electrical infrastructure cannot support the projected growth without substantial investment. Data centers represented 38% of charges in PJM's most recent capacity auction, demonstrating that even amid these challenges, data centers remain the primary driver of new demand.

Some utilities have begun questioning whether they can sustain operations under these conditions. American Electric Power's threat to exit PJM reflects the frustration of legacy utility companies struggling to manage rapid technological change. The situation has created a feedback loop where high prices attract more data center investment, further straining existing infrastructure.

Forecasting Accuracy and Rapid Revisions

The consistent pattern of forecast revisions reveals how quickly the data center expansion is accelerating.

›BloombergNEF's 2035 forecast increased 83% from its December 2023 prediction.
›EPRI more than doubled its 2024 electricity demand estimate.
›S&P increased its forecast by more than one-third between October and April.
›These revisions reflect actual data center development announcements rather than methodological changes.

The pattern of upward forecast revisions across multiple independent organizations indicates that the pace of data center expansion is genuinely accelerating beyond previous expectations. These are not marginal adjustments but substantial increases that suggest forecasters have been systematically underestimating AI computing demand. Each new forecast revision serves as evidence that previous estimates were too conservative, creating a feedback loop where near-term predictions become outdated within months.

The consistency of upward revisions across different forecasting organizations suggests this is not a methodological disagreement but a fundamental change in the trajectory of AI infrastructure investment. When organizations like EPRI double their estimates and S&P increases forecasts by a third, it indicates that major technology companies have accelerated their data center deployment schedules. The revisions reflect actual announcements of facility construction, energy contracts, and equipment orders rather than speculative modeling adjustments.

Global Data Center Expansion Beyond the U.S.

While the U.S. dominates AI compute, data centers will expand globally, creating massive worldwide electricity demands.

›By 2033, new data center demand could reach 1,935 terawatt-hours annually, equivalent to India's total electricity consumption.
›Global expansion will continue despite U.S. concentration of AI infrastructure.
›International markets will develop their own data center capacity as AI adoption spreads.
›The worldwide electricity demand for data centers represents a major component of global energy consumption growth.

The global implications of data center growth extend far beyond the United States. While American grids will bear the immediate strain, countries worldwide are building data center capacity to support local AI deployment and serve as regional hubs. This global expansion could result in 1,935 terawatt-hours of new electricity demand by 2033, a staggering figure that exceeds India's current annual consumption.

The geographic distribution of data center growth will create distinct regional challenges. Countries with abundant renewable energy, like Iceland and Norway, are attracting data center investment. Conversely, regions with tight electricity markets and aging infrastructure face significant challenges meeting new demand. The global nature of AI computing means that decisions made by companies in the U.S. ripple through international markets, influencing energy infrastructure investment worldwide.

Energy Infrastructure Investment Requirements

Meeting projected data center demand will require substantial investments in electrical generation and grid modernization.

›Existing grids cannot support projected data center growth without significant upgrades.
›New generating capacity must be built while data centers continue to connect to the grid.
›Grid congestion has forced utilities to prioritize connections and manage demand dynamically.
›Long-term energy contracts between data center operators and utilities are reshaping the power market.

The infrastructure gap between current grid capacity and projected data center demand creates both challenges and opportunities. Utilities must simultaneously generate additional electricity, upgrade transmission infrastructure, and manage interconnection queues. The situation has attracted significant capital investment, with energy companies prioritizing projects that can serve data center loads.

Major technology companies are increasingly securing power through direct contracts with utilities and renewable energy developers, bypassing traditional market mechanisms. These long-term power purchase agreements lock in prices and availability, but they also reduce flexibility for other electricity consumers. The power market is undergoing a structural shift where a few large buyers of electricity are reshaping supply chains and infrastructure investment priorities.

Climate and Environmental Implications

The massive growth in data center electricity consumption raises significant questions about emissions and sustainability.

›Data center electricity demand could reach one-fifth of U.S. generation by 2035, a enormous component of energy consumption.
›The location of data centers in regions with mixed electricity sources creates emissions challenges.
›Energy efficiency improvements alone cannot offset the growth in absolute electricity demand.
›Renewable energy development must accelerate to meet data center growth without increasing emissions.

The environmental impact of data center expansion depends heavily on how electricity is generated. If new capacity comes primarily from renewable sources, the climate impact could be manageable. However, if coal and natural gas plants continue operating to meet demand, emissions will rise significantly. The concentration of data centers in specific regions means that local air quality and emissions profiles will shift dramatically in affected areas.

Frequently Asked Questions

How much more electricity will data centers use by 2035?

Data centers are expected to use four times more electricity by 2035 compared to today, accounting for one-fifth of all U.S. electricity generation. This massive increase is primarily driven by AI computing demands.

Why have electricity demand forecasts been revised so much?

Multiple forecasting organizations have increased their estimates substantially because major technology companies have accelerated data center construction timelines. BloombergNEF's forecast increased 83% from December, EPRI more than doubled its estimate, and S&P raised its forecast by over a third, reflecting actual announcement of facility builds rather than speculative changes.

Which U.S. regions will be most impacted by data center electricity demand?

PJM Interconnection, spanning Virginia to Illinois, will dedicate 34% of its electricity to data centers and already faces severe grid congestion. ERCOT in Texas will allocate 22% of its generating capacity. PJM has experienced a 76% increase in electricity prices over the past year due to this strain.

Will data center growth remain concentrated in the United States?

While the U.S. will host 64% of global AI chip capacity by power demand by 2033, data centers will continue expanding globally. By 2033, new data center electricity demand worldwide could reach 1,935 terawatt-hours annually, equivalent to India's current total consumption.

What is forcing utilities like American Electric Power to reconsider grid participation?

Grid congestion in regions like PJM has become so severe that managing thousands of new connection requests is overwhelming existing infrastructure. When PJM paused connections for four years, backlogs accumulated substantially, causing some utilities to question whether they can sustain operations under these conditions.

The convergence of explosive data center growth, limited grid capacity, and rising electricity demand represents one of the most significant infrastructure challenges facing the power sector in the coming decade.